We've had stories about the death of networking as we know it for ages now, with every new company in the space being touted as a "Cisco-killer". It's like Chicken Little calling out that the sky is falling; if the blue doesn't hit you, then at some point our chicken loses credibility. But we may really be facing a major sea change in networking, one large enough to kill off vendors and change paradigms forever.
Software-defined networking (SDN) was supposed to kill off switches and routers by replacing their adaptive behavior with centrally controlled operations. We've not gotten much further along in that initiative than the OpenFlow control protocol, and Cisco of all people has jumped into providing an OpenFlow controller in open-source form (Open Daylight). Obviously they're not too afraid of SDN, and they shouldn't be--centralized control just doesn't scale and so it's not a major threat.
What is potentially a major threat is a notion of "hosted routing". Quietly, over the last year or so, both chip vendors and software providers have become more interested in making servers better at networking. Initially the thrust was to better support virtualization and cloud computing, but cloud data center applications showed that virtual networking, a form of SDN, could mean virtual switches and routers.
Some vendors, including Vyatta (bought by Brocade) demonstrated that you could combine a powerful server with a virtual router and actually beat out some real routers in performance. Now we have Cumulus, a stealth startup who has opened up with a plan to create software routers that take advantage of open switches designed in part for SDN, or even just plain servers, for use in building cheap networks.
It's easy to scoff at this idea, particularly when giant Cisco has just come out with a big capacity update for its current core router and is said to be preparing an even bigger box. I don't think anyone believes that white-box or server-based routers are going to replace the purpose-built behemoths.
They don't have to, because network operators have made it clear that they want agile optical layers to do more work in the core. It's nearer the edge, where most of the boxes (and money) are deployed that these software-based routers could have an impact...a major one.
Unlike SDN, the Vyatta or Cumulus approach isn't trying to change the architecture of an IP network, just how it's built out. There's no need to wait for standards to evolve or for issues to be addressed because the functionality of these software routers is nearly identical to the "real" devices (there are always some proprietary extensions and interfaces to deal with). My model says that the software-router model could address almost three-quarters of the router market, and at the least it would put more price pressure on the real routers--and their vendors.
The same server/software development that's creating the software router threat is also facilitating network operator plans to migrate higher-layer network features off appliances and onto commodity servers. The Network Functions Virtualization (NFV) initiative is already well along in defining how to do that, and what I think is the first public seminar on NFV was held in late June--by open-source giant Red Hat together with 6WIND, a vendor of server-network-enhancement software. Even before we have firm specs for NFV, it's clear that the key enabler is the network-optimized server and operating system software.
If servers got fast enough and if the right chips were put into server interfaces, it's possible that networking might become a function of one or more of the virtual machines in each cloud server. It's also possible that things like that centralized SDN control that's been slow in evolving might be sped up if networking were based more on open-source technology, stuff that a community of interested parties would enhance. Even Open Daylight might become a platform for this evolution, which might cause some concern at Cisco, a key sponsor.
There are questions to be addressed in this software-router evolution/revolution, of course. One big one is reliability--the old mean-time-between-failures or MTBF. Purpose-built devices are typically more reliable than servers are. Proponents of software-driven networking models argue that redundancy can address that fundamental difference in device reliability, but failover isn't the same as not failing, particularly when any failure will require reconfiguring routes--"converging," in router terms. It remains to be seen just how a high-availability software-routing model would look or would work.
We have a long way to go before we can say that traditional routers are a dying breed, or that vendors like Cisco are at risk, but we do seem to have reached a point where technical advances in server/software technology will support real competition for routers for the first time. Cisco may have seen that coming, and its remaking of itself into an IT giant may be its way of dealing with the inevitable. So this time, reports of the death of the "old networking" may not be exaggerated.
Follow Tom Nolle on Google+!
Tom Nolle on Google+
